Automatic extraction of social network users' attitudes on reproductive behavior issues

This paper presents a specialized dataset with annotation of user attitudes on reproductive behavior. We analyze the features of the “for” and “against” stance distribution for specific aspects of reproductive behavior. The created dataset solves two classification problems: classifying messages by the relevance to a topic being studied and the author’s stance on a particular issue. We use classical machine learning methods and the BERT-based neural network classified messages models. The best classification results in both tasks are achieved based on variants of the BERT model using pairs of sentences in the classification — variants of NLI (natural language inference) and QA (question-answering). In addition, the created dataset makes it possible to draw meaningful conclusions on the attitudes of VKontakte users to reproductive behavior issues. It was revealed that the phenomenon of deliberate childlessness is actively represented in VKontakte groups while having many children remains a poorly widespread model of behavior. Within the framework of the pro-natalist policy, it is crucial to form a favorable public opinion about parenting, to alleviate the deficiency of time for parents.